In the United Kingdom, traffic accidents led to 1782 deaths and 25,484 serious injuries in 2018. In addition to these direct human costs, crashes, collisions and breakdowns also cause severe congestion leading to significant drops in road efficiency. Hence, there is an imperative to further reduce the accident rate on UK roads. Yet further reductions will likely require targeted interventions to improve safety at specific locations where the accident risk is known to be high, or to mitigate against particular mechanisms known to account for a significant proportion of accidents.
In a new paper, LML External Fellow Colm Connaughton and colleagues contribute to this effort by using traffic data to model the distribution of motorway incidents as a spatiotemporal process comprised of a background component and a self‐excitation component associated with secondary events. These occur when a driver, reacting to the disruption of one accident, triggers a subsequent accident upstream of the first one. The researchers focus on one year of data for the M25 London Orbital, one of the busiest motorways in the United Kingdom. They first quantify how accident risk on the M25 varies in space and time relative to the baseline, and then also estimate the likely contribution of secondary incidents.
Their results indicate that the spatiotemporal variation in the incident rate on the M25 is strongly inhomogeneous, showing a strong daily double peak structure reflecting commuting patterns superimposed in a weaker weekly variation with a peak on Fridays and a trough on Saturdays. This pattern of temporal variation remained stable over the year studied. The spatial variation also showed two primary peaks in intensity, the largest in the vicinity of the Potters Bar Interchange, and the other near Junctions 5 and 6 where the M26 and M23 join the M25. Overall, they found that 6–7% of the observed incidents were likely secondary incidents under the assumptions of their model. They conclude that, on the M25, the scope to further reduce accident rates by reducing secondary incidents is limited, and more could be achieved by reducing the peaks at specific times or ‘hot‐spot’ locations. The authors hope this model will be helpful in addressing the question of how best to target interventions when the baseline accident rate is already low in absolute terms.
The article is available at https://rss.onlinelibrary.wiley.com/doi/full/10.1111/rssc.12450